Starting With the Evidence
A timelapse incubator captures embryo development over several days. Those images contain a detailed record of when developmental stages appear, but interpreting that record remains specialist work performed by trained embryologists.
Our machine-learning research asks a bounded question: can a model identify useful developmental stages consistently enough to support that work? The goal is not to replace clinical judgement. It is to investigate whether carefully evaluated software can provide a structured second view of the timelapse.
Building the Research Foundation
The difficult part is not producing a prediction. It is building evidence that the prediction deserves attention. We are developing a governed dataset and a repeatable evaluation pipeline that keeps training, validation, and future final testing separate.
Experiments are compared against established baselines and rejected when they do not improve the evidence. If a dataset or evaluation check is incomplete, the process stops instead of producing a convenient result.
Privacy by Design
The intended architecture keeps model processing on clinic-controlled hardware alongside TimeLapse Connect. Protecting embryo imagery is therefore a design constraint from the beginning, not a feature to add after a model has been trained.
Where the Work Stands
This remains research. There is no production model and no claim of clinical validation. The current work is focused on dataset quality, reproducible baselines, and evaluating alternative model architectures without relaxing the evidence gates.
Only after those steps succeed would integration into TimeLapse Connect, target-hardware testing, and clinical evaluation become relevant. A model should earn its place in a clinical workflow through evidence, not novelty.